Data Visualization of FBI Crime Data Agencies¶

Author: Srinivasan Poonkundran

Research Question:¶

How do agency types vary across different states in the USA, and what percentage of agencies in each state participate in NIBRS reporting? Are there observable trends in NIBRS adoption over time?

About dataset:¶

We’re looking at agency-level data across all 50 states in the USA. This dataset provides details on law enforcement agencies that have submitted data to the FBI’s Uniform Crime Reporting (UCR) Program and are displayed on the Crime Data Explorer (CDE).

The data we have is an agency level data across all the 50 states in United States. This data sets provides the details on law enforcement agencies that have submitted data to the FBI’s Uniform Crime Reporting Program and are displayed on the Crime Data Explorer.

This dataset is a publicly available and sourced from TidyTuesday. It contains information about agencies that have submitted data to the FBI’s Uniform Crime Reporting (UCR) Program. Key variables include agency name, type, location (latitude/longitude, state, county), and whether the agency participates in the National Incident-Based Reporting System (NIBRS), along with the start date of NIBRS participation.

Link to the dataset: https://github.com/rfordatascience/tidytuesday/tree/main/data/2025/2025-02-18

Data Dictionary¶

agencies.csv¶

Variable Type Description
ori character Unique ID used to identify an agency.
county character The county associated with the agency’s jurisdiction within a state.
latitude double The approximate latitude of the agency.
longitude double The approximate longitude of the agency.
state_abbr character The abbreviated two-letter state code for the agency's location.
state character The full name of the state where the agency is located.
agency_name character The official name of the agency.
agency_type character The type or category of the agency, such as city or county.
is_nibrs logical Indicates whether the agency participates in the FBI’s NIBRS program.
nibrs_start_date character The date on which the agency began reporting data to NIBRS.

Here you can find the github repository of the project: https://github.com/srinivasanpoonkundran/INFO-511-Project

In [23]:
import pandas as pd
import matplotlib.pyplot as plt
import folium
from folium.plugins import MarkerCluster
In [3]:
def count_null_vals(data):
    length = len(data)
    null_counts = data.isnull().sum()
    c = null_counts[null_counts > 0]
    xt = True
    for column, null_count in c.items():
        xt = False
        percent = (null_count/length) * 100
        print(f"'{column}' : {null_count} | {percent:.2f}%")
    if(xt):
        print("No Null values")
In [5]:
agencies = pd.read_csv('data/agencies.csv')
agencies.head()
Out[5]:
ori county latitude longitude state_abbr state agency_name agency_type is_nibrs nibrs_start_date
0 AL0430200 LEE 32.604064 -85.353048 AL Alabama Opelika Police Department City True 2021-01-01
1 AL0430100 LEE 32.608086 -85.475136 AL Alabama Auburn Police Department City True 2020-12-01
2 AL0430000 LEE 32.604064 -85.353048 AL Alabama Lee County Sheriff's Office County True 2012-04-01
3 AL0070100 BIBB 33.015893 -87.127148 AL Alabama Centreville Police Department City True 2020-01-01
4 AL0070000 BIBB 32.939995 -87.116830 AL Alabama Bibb County Sheriff's Office County True 2021-01-01
In [6]:
agencies.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 19166 entries, 0 to 19165
Data columns (total 10 columns):
 #   Column            Non-Null Count  Dtype  
---  ------            --------------  -----  
 0   ori               19166 non-null  object 
 1   county            19166 non-null  object 
 2   latitude          17219 non-null  float64
 3   longitude         17219 non-null  float64
 4   state_abbr        19166 non-null  object 
 5   state             19166 non-null  object 
 6   agency_name       19166 non-null  object 
 7   agency_type       17491 non-null  object 
 8   is_nibrs          19166 non-null  bool   
 9   nibrs_start_date  15105 non-null  object 
dtypes: bool(1), float64(2), object(7)
memory usage: 1.3+ MB
In [7]:
agencies.describe()
Out[7]:
latitude longitude
count 17219.000000 17219.000000
mean 38.431584 -88.965840
std 4.785443 13.303851
min -9.000000 -164.188912
25% 34.933239 -95.268743
50% 39.214480 -86.135794
75% 41.717780 -80.198065
max 69.449343 178.338813

Les us now see if there are any null values in the data.

In [8]:
count_null_vals(agencies)

'latitude' : 1947 | 10.16%
'longitude' : 1947 | 10.16%
'agency_type' : 1675 | 8.74%
'nibrs_start_date' : 4061 | 21.19%

The above shows information on the missing values in the dataset. There are 1947 latitude and longitude data missing from the dataset, which is 10.16% of the data. Column 'agency_type' has 1675 missing values which is 8.74% of the total data. 21.19% of the data in 'nibrs_start_date' column is missing. These missing values will be handled in further section.

In [9]:
agencies['agency_type'].unique()
Out[9]:
array(['City', 'County', 'University or College', nan, 'Other',
       'Other State Agency', 'Tribal', 'State Police', 'Unknown'],
      dtype=object)
In [10]:
agencies['agency_name'].unique()
Out[10]:
array(['Opelika Police Department', 'Auburn Police Department',
       "Lee County Sheriff's Office", ...,
       'Thermopolis Police Department', 'Wyoming Highway Patrol',
       'Wind River Agency'], dtype=object)

Most of the data appears clean, except for missing values. We won’t need extensive data transformation.

Handling Missing Values¶

There are two common approaches:

  1. Delete rows with missing values

    • Simple, but may reduce dataset size.
    • Suitable if missing data is minimal or random.

  2. Fill in missing values

    • Apply appropriate logic based on the variable.
    • Examples:
      • Use mean/median for numerical columns like latitude or longitude.
      • Use mode or "Unknown" for categorical columns like county or agency_type.
In [11]:
total_rows = agencies.shape[0]
missing_rows = agencies[agencies.isnull().any(axis=1)]
num_missing_rows = missing_rows.shape[0]
missing_percentage = (num_missing_rows / total_rows) * 100
print(f"Number of rows with missing values: {num_missing_rows}")
print(f"Percentage of rows with missing values: {missing_percentage:.2f}%")

Number of rows with missing values: 5524
Percentage of rows with missing values: 28.82%

If we use approach 1, we will have to drop 29% of the data which is a huge number. Hence let us go with option 2.

Data Cleaning¶

Handle missing latitude/longitude values by imputing state averages.

In [12]:
#fill missing latitude/longitude with state averages
agencies['latitude'] = agencies.groupby('state_abbr')['latitude'].transform(lambda x: x.fillna(x.mean()))
agencies['longitude'] = agencies.groupby('state_abbr')['longitude'].transform(lambda x: x.fillna(x.mean()))

#remove invalid latitude entries
agencies = agencies[agencies['latitude'] > 0]
In [13]:
count_null_vals(agencies)

'agency_type' : 1675 | 8.74%
'nibrs_start_date' : 4061 | 21.19%

Classify agency types based on names (“Police” = City, “Sheriff” = County, others = Other).

In [14]:
#classify missing agency types
def classify_agency(name):
    if pd.isnull(name):
        return "Other"
    if "Sheriff" in name:
        return "County"
    if "Police" in name:
        return "City"
    return "Other"

agencies['agency_type'] = agencies.apply(
    lambda row: row['agency_type'] if pd.notnull(row['agency_type']) else classify_agency(row['agency_name']), axis=1)
In [15]:
count_null_vals(agencies)

'nibrs_start_date' : 4061 | 21.19%

Fill missing NIBRS start dates with median values.

In [16]:
# Fill missing nibrs_start_date with median
agencies['nibrs_start_date'] = pd.to_datetime(agencies['nibrs_start_date'], errors='coerce')
median_date = agencies['nibrs_start_date'].median()
agencies['nibrs_start_date'] = agencies['nibrs_start_date'].fillna(median_date)
In [17]:
count_null_vals(agencies)

No Null values

We have handled all the missing values.

Exploratory Data Analysis (EDA)¶

Distribution of agency types by state¶

In [18]:
#distribution of agency types by state
agency_counts = agencies.groupby(['state_abbr', 'agency_type']).size().unstack(fill_value=0)
agency_counts.plot(kind='bar', stacked=True, figsize=(15, 7))
plt.title('Distribution of Agency Types by State')
plt.xlabel('State')
plt.ylabel('Number of Agencies')
plt.legend(title='Agency Type')
plt.tight_layout()
plt.show()

Follwoing is what we can see in the Distribution of Agency Types by State chart:

  • Most states have City agencies (like city police departments) as the biggest group.

  • County agencies (like sheriff offices) are the second most common.

  • There’s a smaller number labeled as Other, which could be special agencies or ones not clearly named.

Overall agency type percentage¶

In [19]:
#overall agency type percentage
agency_type_counts = agencies['agency_type'].value_counts()
agency_type_counts.plot(kind='pie', autopct='%1.1f%%', figsize=(7, 7))
plt.title('Overall Distribution of Agency Types')
plt.ylabel('')
plt.show()

Following is what we can see in the Overall Distribution of Agency Types chart:

  • City agencies make up the biggest slice of the pie. That means most law enforcement agencies in the U.S. are city-based, like local police departments.

  • County agencies are the next largest group, usually sheriff's offices serving bigger areas outside cities.

  • The smallest slice is labeled Other. These are special or unique agencies that don’t fit neatly into City or County categories.

When you look at the whole country, you see that law enforcement is mostly run by cities, followed by counties, and a small number fall into other types.

NIBRS participation rates by state¶

In [27]:
#NIBRS participation rates by state
nibrs_participation = agencies.groupby('state_abbr')['is_nibrs'].mean().sort_values(ascending=False) * 100
plt.figure(figsize=(15, 7))
nibrs_participation.plot(kind='bar')
plt.title('NIBRS Participation Rates by State')
plt.xlabel('State')
plt.ylabel('Participation Rate (%)')
plt.tight_layout()
plt.show()

Following is what can we see in the above plot:

  • Each bar shows how many agencies in a state report their crime data using the NIBRS system.

  • Most states have bars that almost reach 100%. That means nearly every agency in those states is using NIBRS.

  • Other states have much shorter bars, meaning fewer agencies are participating.

Trend of NIBRS adoption over time¶

In [25]:
#trend of NIBRS adoption over time
nibrs_over_time = agencies['nibrs_start_date'].dt.year.value_counts().sort_index()
plt.figure(figsize=(12, 6))
nibrs_over_time.plot(kind='line', marker='o', figsize=(8, 6))
plt.title('Trend of NIBRS Adoption Over Time')
plt.xlabel('Year')
plt.ylabel('Number of Agencies Adopting NIBRS')
plt.grid(True)
plt.tight_layout()
plt.show()

Following can be depicted from the above chart

  • The line shows how many agencies started using NIBRS each year.

  • In the early years, not many agencies joined—so the line stays low.

  • As time goes on, the line starts rising, meaning more agencies are adopting NIBRS.

There are a few years where the line jumps up sharply. That tells that something has happened. This could be new policies to get more agencies to report. This shows steady growth, but not all agencies joined at once. It took time, and the adoption happened little by little.

Correlation: Agency type vs NIBRS participation¶

In [22]:
#correlation: Agency type vs NIBRS participation
participation_by_type = agencies.groupby('agency_type')['is_nibrs'].mean() * 100
participation_by_type.plot(kind='bar', figsize=(8, 6))
plt.title('NIBRS Participation by Agency Type')
plt.xlabel('Agency Type')
plt.ylabel('Participation Rate (%)')
plt.tight_layout()
plt.show()

Following is the information aquired from the above plot.

  • Each bar shows the percentage of agencies (by type) that are using NIBRS.

  • City agencies have the tallest bar. That means most city police departments are reporting their crime data through NIBRS.

  • County agencies have a lower bar, so fewer of them are participating compared to city agencies.

  • The Other agencies have the shortest bar, meaning the least participation.

City agencies are more likely to adopt NIBRS, while county and other agencies are slower to catch up.

Map¶

In [24]:
m = folium.Map(location=[39.8283, -98.5795], zoom_start=4)
marker_cluster = MarkerCluster().add_to(m)
for idx, row in agencies.iterrows():
    if pd.notnull(row['latitude']) and pd.notnull(row['longitude']):
        color = "green" if row["is_nibrs"] else "red"  # Green for NIBRS participants, red otherwise
        folium.Marker(
            location=[row["latitude"], row["longitude"]],
            popup=f"{row['agency_name']} ({row['state']})",
            icon=folium.Icon(color=color)
        ).add_to(marker_cluster)
m
Out[24]:
Make this Notebook Trusted to load map: File -> Trust Notebook

This is another cool way to get some clear information on the participation data. The green for NIBRS participants, red otherwise.

Conclusion¶

  • Agency Types Across States:

    • Most states have more City agencies, followed by County agencies, with a smaller number classified as Other.

    • The mix of agency types changes state by state. Some states have more County agencies, especially where there are larger rural areas, while others are dominated by City agencies in urban regions.

  • NIBRS Participation by State:

    • NIBRS participation rates vary widely across states.

    • Some states have nearly 100% participation, showing strong adoption.

    • Other states have lower rates, meaning many agencies still don’t report using NIBRS.

  • Trends Over Time:

    • There is a clear upward trend in NIBRS adoption.

    • Adoption started slow but increased steadily over the years.

    • Certain years saw sharp increases, likely due to efforts or policies encouraging agencies to join.

The agency type differs from state to state. City agencies are the most common overall. NIBRS participation rates are high in some states and lower in others. Over time, NIBRS adoption has grown consistently, with noticeable increases in recent years.